Intelligent Tech Channels MEA Issue 05 | Page 27

FEATURE ENTERPRISE AI
are increasingly reassessing their technology environments through the lens of business continuity and operational resilience.
Following disruptions experienced earlier this year, customers have shown greater interest in moving mission-critical workloads between different environments without interrupting business operations. This includes the ability to migrate applications from public cloud infrastructure back to onpremises environments when necessary. For Red Hat, these requirements reinforce the importance of open hybrid cloud architectures that support application portability across different platforms.
The issue extends directly into enterprise AI strategies, where organisations must balance performance, costs, regulatory requirements and infrastructure availability.
Pickering observes that AI adoption is progressing towards production, although large-scale industrial deployments remain limited.“ I would say enterprise AI adoption has moved beyond the proof-of-concept and pilot stages, although we are not yet seeing implementations at an industrial scale.”
Enterprises are adopting a phased approach, introducing AI into specific functions or customer segments before expanding deployment.
However, infrastructure economics remain a major constraint. Hardware prices,
Adrian Pickering, Regional General Manager, MENA, Red Hat particularly memory costs, have increased significantly, while supply limitations are complicating procurement. These pressures can favour hyperscalers with existing infrastructure capacity, while enterprises seeking to build their own environments face substantial upfront investment.
Skills are another challenge. Organisations need professionals capable
Matt Swalley, Senior Director at VDURA of integrating data, managing infrastructure and translating AI capabilities into reliable production applications.
For channel partners, these developments reinforce the importance of hybrid cloud expertise, application modernisation, infrastructure optimisation and managed services. Customers increasingly need partners capable of helping them make architectural decisions that balance flexibility, resilience and financial sustainability.
Storage emerges as an AI bottleneck The infrastructure debate has largely centred on GPUs, compute capacity and energy availability. However, another component is becoming increasingly important: storage.
Matt Swalley, Senior Director at VDURA, believes storage performance is already a major constraint on AI infrastructure efficiency.“ Storage is already one of the major bottlenecks in AI infrastructure.”
As organisations invest heavily in GPU clusters, the ability to move data rapidly between storage and compute resources becomes essential. Poor storage performance can leave expensive GPUs underutilised, undermining the economics of AI deployments.
VDURA addresses this challenge through a parallel file system designed to support simultaneous data access across compute resources. Its architecture allows performance, capacity and metadata to scale independently, enabling customers to adapt their infrastructure to changing workload requirements.
The company also combines NVMe flash storage with traditional hard disk capacity through context-aware tiering. Data initially lands on flash for high-speed access before being moved automatically to disk according to usage patterns.
According to Swalley, around 90 % of hot data typically remains on flash, while approximately 90 % of overall capacity resides on disk. This approach is intended to combine high performance with more economical storage capacity. Storage efficiency becomes particularly important during AI training and inference.
Checkpointing operations can require large volumes of simultaneous writes, potentially leaving GPUs idle while data is transferred. Inference workloads introduce different requirements, particularly around moving information rapidly between GPU-local storage and high-performance storage tiers.
For organisations investing millions in accelerated computing infrastructure, improving data movement can directly influence infrastructure utilisation and operating costs. The emergence of neocloud providers is also changing the market.
Swalley sees growing demand for specialised GPU cloud environments that allow enterprises to consume AI infrastructure without building everything themselves. For infrastructure distributors and integrators, these developments create opportunities beyond GPU supply, including highperformance storage, networking, architecture design and infrastructure integration. •
The implication for technology partners is significant. Rather than positioning AI primarily around technological capabilities, partners increasingly need to help customers identify suitable use cases, establish performance benchmarks and demonstrate measurable returns.
INTELLIGENT TECH CHANNELS MIDDLE EAST AND AFRICA 27